perf: skip alias loop and BitSet clone for simple eval strings (issue #1961)

docs/about/changelog.md

@@ -42,7 +42,7 @@ priorities and future plans.
 
 - Preserve named typeglob stringification when copying package stash entries (#1927).
 
-- Reduce eval STRING overhead with lazy goto collections, parser-set AST flags that skip unneeded BytecodeCompiler analysis passes, ScopedSymbolTable snapshot caching, per-thread eval caches without synchronization, cached constructor/class-field metadata, and a re-entrant fast-path in `PerlRuntime.bind()` (#1911).
+- Reduce eval STRING overhead with lazy goto collections, parser-set AST flags that skip unneeded BytecodeCompiler analysis passes, ScopedSymbolTable snapshot caching, per-thread eval caches without synchronization, cached constructor/class-field metadata, a re-entrant fast-path in `PerlRuntime.bind()`, and a lexer `sawBeginOrSub` flag that skips the O(visible_vars) outer-lexical alias loop on the common path (#1911, #1961).
 
 - Preserve arbitrary bytes in SQLite text columns across DBI fetches (#1923).
 

src/main/java/org/perlonjava/frontend/lexer/Lexer.java

@@ -73,6 +73,10 @@ public class Lexer {
     public int position;
     // Length of the input
     public int length;
+    // Set to true if the token stream contains BEGIN, sub, use, or no —
+    // keywords that require the outer-lexical alias loop in evalStringWithInterpreter().
+    // False on the common path (simple expressions, while loops, etc.).
+    public boolean sawBeginOrSub = false;
 
     // Constructor to initialize the Lexer with input string
     public Lexer(String input) {
@@ -368,7 +372,11 @@ public LexerToken consumeIdentifier() {
             }
         }
         // Build token text using substring to preserve surrogate pairs correctly
-        return new LexerToken(LexerTokenType.IDENTIFIER, input.substring(start, position));
+        String text = input.substring(start, position);
+        if (!sawBeginOrSub
+                && (text.equals("BEGIN") || text.equals("sub")
+                    || text.equals("use") || text.equals("no"))) sawBeginOrSub = true;
+        return new LexerToken(LexerTokenType.IDENTIFIER, text);
     }
 
     public LexerToken consumeOperator() {

src/main/java/org/perlonjava/frontend/semantic/ScopedSymbolTable.java

@@ -1062,15 +1062,18 @@ public ScopedSymbolTable snapShot(SymbolTable prebuiltTable) {
 
         // Clone warning flags
         st.warningFlagsStack.pop(); // Remove the initial value pushed by enterScope
-        st.warningFlagsStack.push((BitSet) this.warningFlagsStack.peek().clone());
+        BitSet wf = this.warningFlagsStack.peek();
+        st.warningFlagsStack.push(wf.isEmpty() ? new BitSet() : (BitSet) wf.clone());
 
         // Clone disabled warnings flags
         st.warningDisabledStack.pop(); // Remove the initial value pushed by enterScope
-        st.warningDisabledStack.push((BitSet) this.warningDisabledStack.peek().clone());
+        BitSet wd = this.warningDisabledStack.peek();
+        st.warningDisabledStack.push(wd.isEmpty() ? new BitSet() : (BitSet) wd.clone());
 
         // Clone fatal warnings flags
         st.warningFatalStack.pop(); // Remove the initial value pushed by enterScope
-        st.warningFatalStack.push((BitSet) this.warningFatalStack.peek().clone());
+        BitSet wft = this.warningFatalStack.peek();
+        st.warningFatalStack.push(wft.isEmpty() ? new BitSet() : (BitSet) wft.clone());
 
         // Clone feature flags
         st.featureFlagsStack.pop(); // Remove the initial value pushed by enterScope

src/main/java/org/perlonjava/runtime/runtimetypes/RuntimeCode.java

@@ -5393,10 +5393,14 @@ public static RuntimeList evalStringWithInterpreter(
                     : topLevelMain ? runtimePackage : capturedSymbolTable.getCurrentPackage();
 
             try {
+                // Parse the eval string (tokenize first so sawBegin is available for the alias gate below)
+                Lexer lexer = new Lexer(evalString);
+                tokens = lexer.tokenize();
                 // Alias captured outer variables before parsing.
                 // BEGIN/use execute during parsing, so their aliases must be in place first.
                 // Named subs in eval also access outer lexicals via these aliases.
-                {
+                // Skip this O(visible_vars) loop on the common path (no BEGIN or named sub).
+                if (lexer.sawBeginOrSub) {
                 Map<Integer, SymbolTable.SymbolEntry> capturedVars = capturedSymbolTable.getAllVisibleVariables();
                 for (SymbolTable.SymbolEntry entry : capturedVars.values()) {
                     if (!entry.name().equals("@_") && !entry.decl().isEmpty() && !entry.name().startsWith("&")) {
@@ -5452,9 +5456,6 @@ public static RuntimeList evalStringWithInterpreter(
                 }
                 evalCompilerOptions.fileName = evalFilename;
                 warnSignatureArgsInEval(evalString, evalFilename);
-                // Parse the eval string
-                Lexer lexer = new Lexer(evalString);
-                tokens = lexer.tokenize();
                 // Create parser context
                 ScopedSymbolTable parseSymbolTable = capturedSymbolTable.snapShot();
                 parseSymbolTable.setCurrentPackage(